Microsoft pulled the plug on its planned AI-powered history search for Edge this week, striking the feature from the Microsoft 365 Roadmap just days before its expected preview rollout. The cancellation, confirmed through a Roadmap entry update on June 25, 2026, ends one of the browser’s most privacy-forward experiments—an on-device large language model that would scan local browsing history without ever sending data to Microsoft’s cloud.

The move leaves privacy advocates and early testers questioning whether the company’s commitment to local AI processing is wavering, even as competitors double down on similar capabilities.

The feature, internally codenamed “Recall Search” but publicly described as “AI History Search,” was designed to let users find previously visited pages by describing the content in natural language. Instead of traditional keyword matching, the on-device AI would parse sentences like “the article about cat food brands with an orange header” and return accurate results, all while keeping user data strictly on the PC.

A Look Back at the Feature

Microsoft first teased AI History Search at Build 2025, showcasing how Edge could leverage on-device neural processing units (NPUs) and small language models to index and retrieve browsing data without internet connectivity. The feature appeared on the official Microsoft 365 Roadmap in January 2026 with Feature ID 182651, targeting a public preview in June 2026 and a full rollout by August.

During a demonstration, the company emphasized that the AI model—a quantized variant of Phi Silica—would run entirely on the user’s hardware, processing history entries as they were generated and maintaining a local vector database. Queries were executed locally, with zero telemetry sent back to Microsoft servers. This architecture was pitched as a paradigm shift from cloud-reliant AI, addressing growing enterprise and consumer concerns over data sovereignty.

“AI History Search is not optional telemetry, it’s not collection, it’s not anything that leaves your machine,” said a product manager at the Build session. “The index lives encrypted on your disk, the model doesn’t phone home, and you can delete it any time.”

The Privacy Promise

The local-only design distinguished Edge’s approach from Google’s AI-driven history search, which relies on server-side processing and has drawn sharp criticism for its data collection practices. Microsoft positioned the feature as a privacy-respecting alternative, hoping to attract users and organizations uncomfortable with cloud-based AI scanning their browsing habits.

For highly regulated industries—finance, health care, legal—the on-device promise was a key differentiator. Several enterprise customers reportedly began piloting early internal builds, according to forum posts, expecting the feature to simplify knowledge retrieval without compromising compliance.

Yet the very architecture that promised privacy may have also sealed the feature’s fate. Running a language model continuously in the background, indexing each new page, and performing similarity searches on demand proved computationally expensive. Early adopters on Windows-on-Arm devices with Snapdragon X Elite chips reported noticeable battery drain, while x86 laptops without dedicated NPUs often fell back on CPU inference that made fans spin constantly.

What Went Wrong?

The cancellation caught many Insiders off guard. Just days before the June 25 update, the feature was still listed as “In development” with a firm preview timeline. The revised entry now reads “Cancelled” with no explanatory note, a rare move for a feature that had progressed to late-stage testing.

Several factors likely converged to force Microsoft’s hand:

  • Performance and Hardware Constraints: Despite advances in on-device AI, the processor overhead remained unacceptable. Benchmarks leaked to tech forums showed the indexing process consuming upwards of 15% CPU on an Intel Core Ultra 7 155H while idle, triggering thermal throttling on thin-and-light laptops. Even on Snapdragon X Elite, NPU usage spiked to 40% during active browsing sessions, competing with other AI workloads like Windows Studio Effects.
  • Accuracy and Hallucination Risks: Testers reported that the local model occasionally returned fabricated or irrelevant results, a well-known achilles heel of smaller language models. In one incident, a query for “the flight booking I made to London” surfaced a medical appointment confirmation, raising red flags about reliability for mission-critical recall.
  • Shifting AI Strategy: Microsoft has increasingly bet on cloud-connected Copilot experiences across its product line. The company’s earnings call in April 2026 highlighted record Copilot subscription revenue, driven by enterprise upsells. A fully offline AI tool that generates no telemetry offers no hook for subscription attachments or AI credits, making it harder to monetize and justifying the engineering investment.
  • Regulatory Implications: Paradoxically, the local nature of the feature may have introduced legal risks. In jurisdictions with strict data retention policies, having an AI model that can surface sensitive historical data—even if it never leaves the device—could be considered a data processing activity subject to compliance audits. Legal teams might have flagged the risk of an AI agent that effectively “remembers” everything a user has ever seen in Edge, accessible by anyone with physical access to the machine.

Community and Enterprise Reaction

Reaction across developer forums and Reddit threads has been swift and largely negative. “It’s the classic Microsoft move—tease a privacy-first feature to build goodwill, then kill it before it ships,” wrote one user on the Edge Insider subreddit. Another noted, “They just proved that on-device AI is the future, but they aren’t willing to take the short-term performance hit to get there.”

Enterprise IT managers expressed frustration over the reversal. Several had lobbied for the feature as a counter to shadow AI usage, where employees paste sensitive data into cloud-based chatbots. Losing it now leaves them with no effective alternative that combines convenience and data residency.

Privacy advocates pointed out the irony: by cancelling a genuinely local AI tool, Microsoft may push users toward cloud-based solutions from competitors that offer AI history search with far weaker privacy safeguards. Google Chrome’s version, for instance, transmits queries to the cloud and has drawn regulatory scrutiny in the EU.

The Broader AI Landscape

The Edge cancellation fits into a larger pattern of tech companies struggling to balance on-device AI with business models. Apple’s Private Cloud Compute for Apple Intelligence was designed to mitigate this tension by processing sensitive requests on-device first, falling back to encrypted cloud compute only when necessary. Google has been developing Gemini Nano for Chrome, but its implementation remains deeply tied to cloud services for quality results.

Microsoft’s own Copilot+ PCs were marketed around on-device AI, with features like Recall, Cocreate, and Live Captions all running locally. But Recall, in particular, faced a backlash over security when it was revealed that the screenshot databases were stored unencrypted on disk—an oversight Microsoft eventually patched. The controversy may have made the Edge team wary of shipping a similar “local but sensitive” feature, especially one that indexes the full text of every page visited.

What’s Next for Edge AI Features?

While AI History Search is dead, Microsoft has not abandoned all local AI ambitions in Edge. The browser still offers on-device text prediction, a local translation engine powered by the Microsoft Translator model, and a basic NPU-driven auto-captioning tool for media. The company is also experimenting with a cloud-assisted history aggregation feature under the “Copilot in Edge” umbrella, which would upload anonymized usage patterns to help users discover content across devices.

In a statement provided to Windows News, a Microsoft spokesperson said: “We constantly evaluate our features to ensure they meet customer needs and our high performance standards. After thorough testing, we decided not to ship AI History Search in its current form. We remain committed to bringing innovative, privacy-respecting AI capabilities to Edge and will share more later this year.”

The statement leaves the door open for a reimagined version—perhaps one that offloads heavier processing to Microsoft servers while encrypting data in transit and at rest, or a tiered approach where basic search runs locally and complex semantic queries leverage the cloud with explicit consent.

Conclusion: A Cautionary Tale for Local AI

The swift rise and fall of Edge’s AI History Search illustrates the massive gap between the promise of on-device AI and the reality of shipping it. Users want instantaneous, intelligent local search that never leaks their data. Hardware makers are shipping NPUs that claim to deliver it. But the software still struggles to balance performance, accuracy, and battery life without compromise.

For Windows enthusiasts and enterprise IT decision-makers, the takeaway is clear: don’t bank on roadmap features until they’re actually installed. Microsoft’s pivot also suggests that truly local AI will remain a niche, at least until the next generation of low-power accelerators arrives. Until then, your browser history will stay dumb—and ironically, that might be the smartest privacy move of all.